REU Site: Hands-On Research in Federated Learning Security through Red Team vs. Blue Team Exercises

This Research Experiences for Undergraduates Site at the University of Nevada, Las Vegas supports 10 students each year in a 10-week summer research program on the security of federated learning, a way for many devices or organizations to train a shared artificial intelligence model without exchanging their raw data. This approach can help protect privacy, but it also creates new security risks because attackers may try to corrupt the training process, steal information from the model, or reduce system reliability. The project’s novelties are the integration of hands-on attack-and-defense research across the full federated learning process and the use of Red Team versus Blue Team exercises to study these problems in realistic settings. The project's broader significance and importance are that it advances safer privacy-preserving artificial intelligence, expands access to advanced undergraduate research opportunities, and helps prepare the future artificial intelligence and cybersecurity workforce. The project contributes to a stronger national capacity for building trustworthy data-driven systems. The research project focuses on threats and defenses in the data collection, training, and inference stages of federated learning. Students and mentors investigate representative attacks including botnet-style disruption, poisoning, backdoor insertion, privacy leakage, membership inference, and data reconstruction, and they evaluate defenses such as robust aggregation, anomaly detection, and differential privacy. The work uses a dedicated federated learning cybersecurity range, realistic datasets from computer vision, language, and network traffic applications, and distributed computing resources for controlled experimentation. Through iterative Red Team and Blue Team studies, the project produces software, tutorials, datasets, and empirical results that improve understanding of secure and privacy-preserving distributed learning. The anticipated outcome is stronger technical foundations for trustworthy artificial intelligence and a broader pipeline of students prepared for research and professional practice in cybersecurity and artificial intelligence. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.